4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
25 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2012 Pawel Jakub Dawidek. All rights reserved.
27 * Copyright (c) 2013 Steven Hartland. All rights reserved.
28 * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
30 * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>
43 #include <sys/param.h>
44 #include <sys/mount.h>
50 #include <libzfs_core.h>
52 #include "zfs_namecheck.h"
54 #include "zfs_fletcher.h"
55 #include "libzfs_impl.h"
58 #include <sys/zio_checksum.h>
62 extern int zfs_ioctl_version;
65 /* in libzfs_dataset.c */
66 extern void zfs_setprop_error(libzfs_handle_t *, zfs_prop_t, int, char *);
67 /* We need to use something for ENODATA. */
70 static int zfs_receive_impl(libzfs_handle_t *, const char *, const char *,
71 recvflags_t *, int, const char *, nvlist_t *, avl_tree_t *, char **, int,
72 uint64_t *, const char *);
73 static int guid_to_name(libzfs_handle_t *, const char *,
74 uint64_t, boolean_t, char *);
76 static const zio_cksum_t zero_cksum = { 0 };
78 typedef struct dedup_arg {
81 libzfs_handle_t *dedup_hdl;
84 typedef struct progress_arg {
87 boolean_t pa_parsable;
90 typedef struct dataref {
96 typedef struct dedup_entry {
97 struct dedup_entry *dde_next;
98 zio_cksum_t dde_chksum;
103 #define MAX_DDT_PHYSMEM_PERCENT 20
104 #define SMALLEST_POSSIBLE_MAX_DDT_MB 128
106 typedef struct dedup_table {
107 dedup_entry_t **dedup_hash_array;
108 umem_cache_t *ddecache;
109 uint64_t max_ddt_size; /* max dedup table size in bytes */
110 uint64_t cur_ddt_size; /* current dedup table size in bytes */
117 high_order_bit(uint64_t n)
121 for (count = 0; n != 0; count++)
127 ssread(void *buf, size_t len, FILE *stream)
131 if ((outlen = fread(buf, len, 1, stream)) == 0)
138 ddt_hash_append(libzfs_handle_t *hdl, dedup_table_t *ddt, dedup_entry_t **ddepp,
139 zio_cksum_t *cs, uint64_t prop, dataref_t *dr)
143 if (ddt->cur_ddt_size >= ddt->max_ddt_size) {
144 if (ddt->ddt_full == B_FALSE) {
145 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
146 "Dedup table full. Deduplication will continue "
147 "with existing table entries"));
148 ddt->ddt_full = B_TRUE;
153 if ((dde = umem_cache_alloc(ddt->ddecache, UMEM_DEFAULT))
155 assert(*ddepp == NULL);
156 dde->dde_next = NULL;
157 dde->dde_chksum = *cs;
158 dde->dde_prop = prop;
161 ddt->cur_ddt_size += sizeof (dedup_entry_t);
167 * Using the specified dedup table, do a lookup for an entry with
168 * the checksum cs. If found, return the block's reference info
169 * in *dr. Otherwise, insert a new entry in the dedup table, using
170 * the reference information specified by *dr.
172 * return value: true - entry was found
173 * false - entry was not found
176 ddt_update(libzfs_handle_t *hdl, dedup_table_t *ddt, zio_cksum_t *cs,
177 uint64_t prop, dataref_t *dr)
180 dedup_entry_t **ddepp;
182 hashcode = BF64_GET(cs->zc_word[0], 0, ddt->numhashbits);
184 for (ddepp = &(ddt->dedup_hash_array[hashcode]); *ddepp != NULL;
185 ddepp = &((*ddepp)->dde_next)) {
186 if (ZIO_CHECKSUM_EQUAL(((*ddepp)->dde_chksum), *cs) &&
187 (*ddepp)->dde_prop == prop) {
188 *dr = (*ddepp)->dde_ref;
192 ddt_hash_append(hdl, ddt, ddepp, cs, prop, dr);
197 dump_record(dmu_replay_record_t *drr, void *payload, int payload_len,
198 zio_cksum_t *zc, int outfd)
200 ASSERT3U(offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum),
201 ==, sizeof (dmu_replay_record_t) - sizeof (zio_cksum_t));
202 fletcher_4_incremental_native(drr,
203 offsetof(dmu_replay_record_t, drr_u.drr_checksum.drr_checksum), zc);
204 if (drr->drr_type != DRR_BEGIN) {
205 ASSERT(ZIO_CHECKSUM_IS_ZERO(&drr->drr_u.
206 drr_checksum.drr_checksum));
207 drr->drr_u.drr_checksum.drr_checksum = *zc;
209 fletcher_4_incremental_native(&drr->drr_u.drr_checksum.drr_checksum,
210 sizeof (zio_cksum_t), zc);
211 if (write(outfd, drr, sizeof (*drr)) == -1)
213 if (payload_len != 0) {
214 fletcher_4_incremental_native(payload, payload_len, zc);
215 if (write(outfd, payload, payload_len) == -1)
222 * This function is started in a separate thread when the dedup option
223 * has been requested. The main send thread determines the list of
224 * snapshots to be included in the send stream and makes the ioctl calls
225 * for each one. But instead of having the ioctl send the output to the
226 * the output fd specified by the caller of zfs_send()), the
227 * ioctl is told to direct the output to a pipe, which is read by the
228 * alternate thread running THIS function. This function does the
230 * 1. building a dedup table (the DDT)
231 * 2. doing checksums on each data block and inserting a record in the DDT
232 * 3. looking for matching checksums, and
233 * 4. sending a DRR_WRITE_BYREF record instead of a write record whenever
234 * a duplicate block is found.
235 * The output of this function then goes to the output fd requested
236 * by the caller of zfs_send().
241 dedup_arg_t *dda = arg;
242 char *buf = zfs_alloc(dda->dedup_hdl, SPA_MAXBLOCKSIZE);
243 dmu_replay_record_t thedrr;
244 dmu_replay_record_t *drr = &thedrr;
248 zio_cksum_t stream_cksum;
249 uint64_t physmem = sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE);
253 MAX((physmem * MAX_DDT_PHYSMEM_PERCENT) / 100,
254 SMALLEST_POSSIBLE_MAX_DDT_MB << 20);
256 numbuckets = ddt.max_ddt_size / (sizeof (dedup_entry_t));
259 * numbuckets must be a power of 2. Increase number to
260 * a power of 2 if necessary.
262 if (!ISP2(numbuckets))
263 numbuckets = 1 << high_order_bit(numbuckets);
265 ddt.dedup_hash_array = calloc(numbuckets, sizeof (dedup_entry_t *));
266 ddt.ddecache = umem_cache_create("dde", sizeof (dedup_entry_t), 0,
267 NULL, NULL, NULL, NULL, NULL, 0);
268 ddt.cur_ddt_size = numbuckets * sizeof (dedup_entry_t *);
269 ddt.numhashbits = high_order_bit(numbuckets) - 1;
270 ddt.ddt_full = B_FALSE;
272 outfd = dda->outputfd;
273 ofp = fdopen(dda->inputfd, "r");
274 while (ssread(drr, sizeof (*drr), ofp) != 0) {
276 switch (drr->drr_type) {
279 struct drr_begin *drrb = &drr->drr_u.drr_begin;
282 ZIO_SET_CHECKSUM(&stream_cksum, 0, 0, 0, 0);
284 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
286 /* set the DEDUP feature flag for this stream */
287 fflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
288 fflags |= (DMU_BACKUP_FEATURE_DEDUP |
289 DMU_BACKUP_FEATURE_DEDUPPROPS);
290 DMU_SET_FEATUREFLAGS(drrb->drr_versioninfo, fflags);
292 if (drr->drr_payloadlen != 0) {
293 sz = drr->drr_payloadlen;
295 if (sz > SPA_MAXBLOCKSIZE) {
296 buf = zfs_realloc(dda->dedup_hdl, buf,
297 SPA_MAXBLOCKSIZE, sz);
299 (void) ssread(buf, sz, ofp);
303 if (dump_record(drr, buf, sz, &stream_cksum,
311 struct drr_end *drre = &drr->drr_u.drr_end;
312 /* use the recalculated checksum */
313 drre->drr_checksum = stream_cksum;
314 if (dump_record(drr, NULL, 0, &stream_cksum,
322 struct drr_object *drro = &drr->drr_u.drr_object;
323 if (drro->drr_bonuslen > 0) {
325 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
328 if (dump_record(drr, buf,
329 P2ROUNDUP((uint64_t)drro->drr_bonuslen, 8),
330 &stream_cksum, outfd) != 0)
337 struct drr_spill *drrs = &drr->drr_u.drr_spill;
338 (void) ssread(buf, drrs->drr_length, ofp);
339 if (dump_record(drr, buf, drrs->drr_length,
340 &stream_cksum, outfd) != 0)
345 case DRR_FREEOBJECTS:
347 if (dump_record(drr, NULL, 0, &stream_cksum,
355 struct drr_write *drrw = &drr->drr_u.drr_write;
358 (void) ssread(buf, drrw->drr_length, ofp);
361 * Use the existing checksum if it's dedup-capable,
362 * else calculate a SHA256 checksum for it.
365 if (ZIO_CHECKSUM_EQUAL(drrw->drr_key.ddk_cksum,
367 !DRR_IS_DEDUP_CAPABLE(drrw->drr_checksumflags)) {
369 zio_cksum_t tmpsha256;
372 SHA256Update(&ctx, buf, drrw->drr_length);
373 SHA256Final(&tmpsha256, &ctx);
374 drrw->drr_key.ddk_cksum.zc_word[0] =
375 BE_64(tmpsha256.zc_word[0]);
376 drrw->drr_key.ddk_cksum.zc_word[1] =
377 BE_64(tmpsha256.zc_word[1]);
378 drrw->drr_key.ddk_cksum.zc_word[2] =
379 BE_64(tmpsha256.zc_word[2]);
380 drrw->drr_key.ddk_cksum.zc_word[3] =
381 BE_64(tmpsha256.zc_word[3]);
382 drrw->drr_checksumtype = ZIO_CHECKSUM_SHA256;
383 drrw->drr_checksumflags = DRR_CHECKSUM_DEDUP;
386 dataref.ref_guid = drrw->drr_toguid;
387 dataref.ref_object = drrw->drr_object;
388 dataref.ref_offset = drrw->drr_offset;
390 if (ddt_update(dda->dedup_hdl, &ddt,
391 &drrw->drr_key.ddk_cksum, drrw->drr_key.ddk_prop,
393 dmu_replay_record_t wbr_drr = {0};
394 struct drr_write_byref *wbr_drrr =
395 &wbr_drr.drr_u.drr_write_byref;
397 /* block already present in stream */
398 wbr_drr.drr_type = DRR_WRITE_BYREF;
400 wbr_drrr->drr_object = drrw->drr_object;
401 wbr_drrr->drr_offset = drrw->drr_offset;
402 wbr_drrr->drr_length = drrw->drr_length;
403 wbr_drrr->drr_toguid = drrw->drr_toguid;
404 wbr_drrr->drr_refguid = dataref.ref_guid;
405 wbr_drrr->drr_refobject =
407 wbr_drrr->drr_refoffset =
410 wbr_drrr->drr_checksumtype =
411 drrw->drr_checksumtype;
412 wbr_drrr->drr_checksumflags =
413 drrw->drr_checksumtype;
414 wbr_drrr->drr_key.ddk_cksum =
415 drrw->drr_key.ddk_cksum;
416 wbr_drrr->drr_key.ddk_prop =
417 drrw->drr_key.ddk_prop;
419 if (dump_record(&wbr_drr, NULL, 0,
420 &stream_cksum, outfd) != 0)
423 /* block not previously seen */
424 if (dump_record(drr, buf, drrw->drr_length,
425 &stream_cksum, outfd) != 0)
431 case DRR_WRITE_EMBEDDED:
433 struct drr_write_embedded *drrwe =
434 &drr->drr_u.drr_write_embedded;
436 P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), ofp);
437 if (dump_record(drr, buf,
438 P2ROUNDUP((uint64_t)drrwe->drr_psize, 8),
439 &stream_cksum, outfd) != 0)
446 if (dump_record(drr, NULL, 0, &stream_cksum,
453 (void) fprintf(stderr, "INVALID record type 0x%x\n",
455 /* should never happen, so assert */
460 umem_cache_destroy(ddt.ddecache);
461 free(ddt.dedup_hash_array);
469 * Routines for dealing with the AVL tree of fs-nvlists
471 typedef struct fsavl_node {
479 fsavl_compare(const void *arg1, const void *arg2)
481 const fsavl_node_t *fn1 = arg1;
482 const fsavl_node_t *fn2 = arg2;
484 if (fn1->fn_guid > fn2->fn_guid)
486 else if (fn1->fn_guid < fn2->fn_guid)
493 * Given the GUID of a snapshot, find its containing filesystem and
497 fsavl_find(avl_tree_t *avl, uint64_t snapguid, char **snapname)
499 fsavl_node_t fn_find;
502 fn_find.fn_guid = snapguid;
504 fn = avl_find(avl, &fn_find, NULL);
507 *snapname = fn->fn_snapname;
508 return (fn->fn_nvfs);
514 fsavl_destroy(avl_tree_t *avl)
523 while ((fn = avl_destroy_nodes(avl, &cookie)) != NULL)
530 * Given an nvlist, produce an avl tree of snapshots, ordered by guid
533 fsavl_create(nvlist_t *fss)
536 nvpair_t *fselem = NULL;
538 if ((fsavl = malloc(sizeof (avl_tree_t))) == NULL)
541 avl_create(fsavl, fsavl_compare, sizeof (fsavl_node_t),
542 offsetof(fsavl_node_t, fn_node));
544 while ((fselem = nvlist_next_nvpair(fss, fselem)) != NULL) {
545 nvlist_t *nvfs, *snaps;
546 nvpair_t *snapelem = NULL;
548 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
549 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
552 nvlist_next_nvpair(snaps, snapelem)) != NULL) {
556 VERIFY(0 == nvpair_value_uint64(snapelem, &guid));
557 if ((fn = malloc(sizeof (fsavl_node_t))) == NULL) {
558 fsavl_destroy(fsavl);
562 fn->fn_snapname = nvpair_name(snapelem);
566 * Note: if there are multiple snaps with the
567 * same GUID, we ignore all but one.
569 if (avl_find(fsavl, fn, NULL) == NULL)
580 * Routines for dealing with the giant nvlist of fs-nvlists, etc.
582 typedef struct send_data {
584 * assigned inside every recursive call,
585 * restored from *_save on return:
587 * guid of fromsnap snapshot in parent dataset
588 * txg of fromsnap snapshot in current dataset
589 * txg of tosnap snapshot in current dataset
592 uint64_t parent_fromsnap_guid;
593 uint64_t fromsnap_txg;
596 /* the nvlists get accumulated during depth-first traversal */
597 nvlist_t *parent_snaps;
601 /* send-receive configuration, does not change during traversal */
603 const char *fromsnap;
609 * The header nvlist is of the following format:
612 * "fromsnap" -> string (if incremental)
616 * "name" -> string (full name; for debugging)
617 * "parentfromsnap" -> number (guid of fromsnap in parent)
619 * "props" -> { name -> value (only if set here) }
620 * "snaps" -> { name (lastname) -> number (guid) }
621 * "snapprops" -> { name (lastname) -> { name -> value } }
623 * "origin" -> number (guid) (if clone)
624 * "sent" -> boolean (not on-disk)
632 static void send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv);
635 send_iterate_snap(zfs_handle_t *zhp, void *arg)
637 send_data_t *sd = arg;
638 uint64_t guid = zhp->zfs_dmustats.dds_guid;
639 uint64_t txg = zhp->zfs_dmustats.dds_creation_txg;
643 snapname = strrchr(zhp->zfs_name, '@')+1;
645 if (sd->tosnap_txg != 0 && txg > sd->tosnap_txg) {
647 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
648 "skipping snapshot %s because it was created "
649 "after the destination snapshot (%s)\n"),
650 zhp->zfs_name, sd->tosnap);
656 VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
658 * NB: if there is no fromsnap here (it's a newly created fs in
659 * an incremental replication), we will substitute the tosnap.
661 if ((sd->fromsnap && strcmp(snapname, sd->fromsnap) == 0) ||
662 (sd->parent_fromsnap_guid == 0 && sd->tosnap &&
663 strcmp(snapname, sd->tosnap) == 0)) {
664 sd->parent_fromsnap_guid = guid;
667 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
668 send_iterate_prop(zhp, nv);
669 VERIFY(0 == nvlist_add_nvlist(sd->snapprops, snapname, nv));
677 send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
679 nvpair_t *elem = NULL;
681 while ((elem = nvlist_next_nvpair(zhp->zfs_props, elem)) != NULL) {
682 char *propname = nvpair_name(elem);
683 zfs_prop_t prop = zfs_name_to_prop(propname);
686 if (!zfs_prop_user(propname)) {
688 * Realistically, this should never happen. However,
689 * we want the ability to add DSL properties without
690 * needing to make incompatible version changes. We
691 * need to ignore unknown properties to allow older
692 * software to still send datasets containing these
693 * properties, with the unknown properties elided.
695 if (prop == ZPROP_INVAL)
698 if (zfs_prop_readonly(prop))
702 verify(nvpair_value_nvlist(elem, &propnv) == 0);
703 if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
704 prop == ZFS_PROP_REFQUOTA ||
705 prop == ZFS_PROP_REFRESERVATION) {
708 verify(nvlist_lookup_uint64(propnv,
709 ZPROP_VALUE, &value) == 0);
710 if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
713 * May have no source before SPA_VERSION_RECVD_PROPS,
714 * but is still modifiable.
716 if (nvlist_lookup_string(propnv,
717 ZPROP_SOURCE, &source) == 0) {
718 if ((strcmp(source, zhp->zfs_name) != 0) &&
720 ZPROP_SOURCE_VAL_RECVD) != 0))
725 if (nvlist_lookup_string(propnv,
726 ZPROP_SOURCE, &source) != 0)
728 if ((strcmp(source, zhp->zfs_name) != 0) &&
729 (strcmp(source, ZPROP_SOURCE_VAL_RECVD) != 0))
733 if (zfs_prop_user(propname) ||
734 zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
736 verify(nvlist_lookup_string(propnv,
737 ZPROP_VALUE, &value) == 0);
738 VERIFY(0 == nvlist_add_string(nv, propname, value));
741 verify(nvlist_lookup_uint64(propnv,
742 ZPROP_VALUE, &value) == 0);
743 VERIFY(0 == nvlist_add_uint64(nv, propname, value));
749 * returns snapshot creation txg
750 * and returns 0 if the snapshot does not exist
753 get_snap_txg(libzfs_handle_t *hdl, const char *fs, const char *snap)
755 char name[ZFS_MAX_DATASET_NAME_LEN];
758 if (fs == NULL || fs[0] == '\0' || snap == NULL || snap[0] == '\0')
761 (void) snprintf(name, sizeof (name), "%s@%s", fs, snap);
762 if (zfs_dataset_exists(hdl, name, ZFS_TYPE_SNAPSHOT)) {
763 zfs_handle_t *zhp = zfs_open(hdl, name, ZFS_TYPE_SNAPSHOT);
765 txg = zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG);
774 * recursively generate nvlists describing datasets. See comment
775 * for the data structure send_data_t above for description of contents
779 send_iterate_fs(zfs_handle_t *zhp, void *arg)
781 send_data_t *sd = arg;
784 uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
785 uint64_t fromsnap_txg_save = sd->fromsnap_txg;
786 uint64_t tosnap_txg_save = sd->tosnap_txg;
787 uint64_t txg = zhp->zfs_dmustats.dds_creation_txg;
788 uint64_t guid = zhp->zfs_dmustats.dds_guid;
789 uint64_t fromsnap_txg, tosnap_txg;
792 fromsnap_txg = get_snap_txg(zhp->zfs_hdl, zhp->zfs_name, sd->fromsnap);
793 if (fromsnap_txg != 0)
794 sd->fromsnap_txg = fromsnap_txg;
796 tosnap_txg = get_snap_txg(zhp->zfs_hdl, zhp->zfs_name, sd->tosnap);
798 sd->tosnap_txg = tosnap_txg;
801 * on the send side, if the current dataset does not have tosnap,
802 * perform two additional checks:
804 * - skip sending the current dataset if it was created later than
806 * - return error if the current dataset was created earlier than
809 if (sd->tosnap != NULL && tosnap_txg == 0) {
810 if (sd->tosnap_txg != 0 && txg > sd->tosnap_txg) {
812 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
813 "skipping dataset %s: snapshot %s does "
814 "not exist\n"), zhp->zfs_name, sd->tosnap);
817 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
818 "cannot send %s@%s%s: snapshot %s@%s does not "
819 "exist\n"), sd->fsname, sd->tosnap, sd->recursive ?
820 dgettext(TEXT_DOMAIN, " recursively") : "",
821 zhp->zfs_name, sd->tosnap);
827 VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
828 VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
829 VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
830 sd->parent_fromsnap_guid));
832 if (zhp->zfs_dmustats.dds_origin[0]) {
833 zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
834 zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
835 if (origin == NULL) {
839 VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
840 origin->zfs_dmustats.dds_guid));
843 /* iterate over props */
844 VERIFY(0 == nvlist_alloc(&nv, NV_UNIQUE_NAME, 0));
845 send_iterate_prop(zhp, nv);
846 VERIFY(0 == nvlist_add_nvlist(nvfs, "props", nv));
849 /* iterate over snaps, and set sd->parent_fromsnap_guid */
850 sd->parent_fromsnap_guid = 0;
851 VERIFY(0 == nvlist_alloc(&sd->parent_snaps, NV_UNIQUE_NAME, 0));
852 VERIFY(0 == nvlist_alloc(&sd->snapprops, NV_UNIQUE_NAME, 0));
853 (void) zfs_iter_snapshots_sorted(zhp, send_iterate_snap, sd);
854 VERIFY(0 == nvlist_add_nvlist(nvfs, "snaps", sd->parent_snaps));
855 VERIFY(0 == nvlist_add_nvlist(nvfs, "snapprops", sd->snapprops));
856 nvlist_free(sd->parent_snaps);
857 nvlist_free(sd->snapprops);
859 /* add this fs to nvlist */
860 (void) snprintf(guidstring, sizeof (guidstring),
861 "0x%llx", (longlong_t)guid);
862 VERIFY(0 == nvlist_add_nvlist(sd->fss, guidstring, nvfs));
865 /* iterate over children */
867 rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
870 sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
871 sd->fromsnap_txg = fromsnap_txg_save;
872 sd->tosnap_txg = tosnap_txg_save;
879 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
880 const char *tosnap, boolean_t recursive, boolean_t verbose,
881 nvlist_t **nvlp, avl_tree_t **avlp)
884 send_data_t sd = { 0 };
887 zhp = zfs_open(hdl, fsname, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
889 return (EZFS_BADTYPE);
891 VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
893 sd.fromsnap = fromsnap;
895 sd.recursive = recursive;
896 sd.verbose = verbose;
898 if ((error = send_iterate_fs(zhp, &sd)) != 0) {
906 if (avlp != NULL && (*avlp = fsavl_create(sd.fss)) == NULL) {
917 * Routines specific to "zfs send"
919 typedef struct send_dump_data {
920 /* these are all just the short snapname (the part after the @) */
921 const char *fromsnap;
923 char prevsnap[ZFS_MAX_DATASET_NAME_LEN];
924 uint64_t prevsnap_obj;
925 boolean_t seenfrom, seento, replicate, doall, fromorigin;
926 boolean_t verbose, dryrun, parsable, progress, embed_data, std_out;
927 boolean_t large_block;
933 snapfilter_cb_t *filter_cb;
936 char holdtag[ZFS_MAX_DATASET_NAME_LEN];
942 estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj,
943 boolean_t fromorigin, uint64_t *sizep)
945 zfs_cmd_t zc = { 0 };
946 libzfs_handle_t *hdl = zhp->zfs_hdl;
948 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
949 assert(fromsnap_obj == 0 || !fromorigin);
951 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
952 zc.zc_obj = fromorigin;
953 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
954 zc.zc_fromobj = fromsnap_obj;
955 zc.zc_guid = 1; /* estimate flag */
957 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
959 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
960 "warning: cannot estimate space for '%s'"), zhp->zfs_name);
964 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
965 "not an earlier snapshot from the same fs"));
966 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
969 if (zfs_dataset_exists(hdl, zc.zc_name,
970 ZFS_TYPE_SNAPSHOT)) {
971 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
972 "incremental source (@%s) does not exist"),
975 return (zfs_error(hdl, EZFS_NOENT, errbuf));
987 zfs_error_aux(hdl, strerror(errno));
988 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
991 return (zfs_standard_error(hdl, errno, errbuf));
995 *sizep = zc.zc_objset_type;
1001 * Dumps a backup of the given snapshot (incremental from fromsnap if it's not
1002 * NULL) to the file descriptor specified by outfd.
1005 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
1006 boolean_t fromorigin, int outfd, enum lzc_send_flags flags,
1009 zfs_cmd_t zc = { 0 };
1010 libzfs_handle_t *hdl = zhp->zfs_hdl;
1013 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
1014 assert(fromsnap_obj == 0 || !fromorigin);
1016 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1017 zc.zc_cookie = outfd;
1018 zc.zc_obj = fromorigin;
1019 zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1020 zc.zc_fromobj = fromsnap_obj;
1021 zc.zc_flags = flags;
1023 VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
1024 if (fromsnap && fromsnap[0] != '\0') {
1025 VERIFY(0 == nvlist_add_string(thisdbg,
1026 "fromsnap", fromsnap));
1029 if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SEND, &zc) != 0) {
1031 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1032 "warning: cannot send '%s'"), zhp->zfs_name);
1034 VERIFY(0 == nvlist_add_uint64(thisdbg, "error", errno));
1036 VERIFY(0 == nvlist_add_nvlist(debugnv,
1037 zhp->zfs_name, thisdbg));
1039 nvlist_free(thisdbg);
1043 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1044 "not an earlier snapshot from the same fs"));
1045 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
1048 if (zfs_dataset_exists(hdl, zc.zc_name,
1049 ZFS_TYPE_SNAPSHOT)) {
1050 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1051 "incremental source (@%s) does not exist"),
1054 return (zfs_error(hdl, EZFS_NOENT, errbuf));
1069 zfs_error_aux(hdl, strerror(errno));
1070 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
1073 return (zfs_standard_error(hdl, errno, errbuf));
1078 VERIFY(0 == nvlist_add_nvlist(debugnv, zhp->zfs_name, thisdbg));
1079 nvlist_free(thisdbg);
1085 gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd)
1087 assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
1090 * zfs_send() only sets snapholds for sends that need them,
1091 * e.g. replication and doall.
1093 if (sdd->snapholds == NULL)
1096 fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag);
1100 send_progress_thread(void *arg)
1102 progress_arg_t *pa = arg;
1103 zfs_cmd_t zc = { 0 };
1104 zfs_handle_t *zhp = pa->pa_zhp;
1105 libzfs_handle_t *hdl = zhp->zfs_hdl;
1106 unsigned long long bytes;
1111 (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1113 if (!pa->pa_parsable)
1114 (void) fprintf(stderr, "TIME SENT SNAPSHOT\n");
1117 * Print the progress from ZFS_IOC_SEND_PROGRESS every second.
1122 zc.zc_cookie = pa->pa_fd;
1123 if (zfs_ioctl(hdl, ZFS_IOC_SEND_PROGRESS, &zc) != 0)
1124 return ((void *)-1);
1128 bytes = zc.zc_cookie;
1130 if (pa->pa_parsable) {
1131 (void) fprintf(stderr, "%02d:%02d:%02d\t%llu\t%s\n",
1132 tm->tm_hour, tm->tm_min, tm->tm_sec,
1133 bytes, zhp->zfs_name);
1135 zfs_nicenum(bytes, buf, sizeof (buf));
1136 (void) fprintf(stderr, "%02d:%02d:%02d %5s %s\n",
1137 tm->tm_hour, tm->tm_min, tm->tm_sec,
1138 buf, zhp->zfs_name);
1144 send_print_verbose(FILE *fout, const char *tosnap, const char *fromsnap,
1145 uint64_t size, boolean_t parsable)
1148 if (fromsnap != NULL) {
1149 (void) fprintf(fout, "incremental\t%s\t%s",
1152 (void) fprintf(fout, "full\t%s",
1156 if (fromsnap != NULL) {
1157 if (strchr(fromsnap, '@') == NULL &&
1158 strchr(fromsnap, '#') == NULL) {
1159 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1160 "send from @%s to %s"),
1163 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1164 "send from %s to %s"),
1168 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1176 (void) fprintf(fout, "\t%llu",
1180 zfs_nicenum(size, buf, sizeof (buf));
1181 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1182 " estimated size is %s"), buf);
1185 (void) fprintf(fout, "\n");
1189 dump_snapshot(zfs_handle_t *zhp, void *arg)
1191 send_dump_data_t *sdd = arg;
1192 progress_arg_t pa = { 0 };
1196 boolean_t isfromsnap, istosnap, fromorigin;
1197 boolean_t exclude = B_FALSE;
1198 FILE *fout = sdd->std_out ? stdout : stderr;
1201 thissnap = strchr(zhp->zfs_name, '@') + 1;
1202 isfromsnap = (sdd->fromsnap != NULL &&
1203 strcmp(sdd->fromsnap, thissnap) == 0);
1205 if (!sdd->seenfrom && isfromsnap) {
1206 gather_holds(zhp, sdd);
1207 sdd->seenfrom = B_TRUE;
1208 (void) strcpy(sdd->prevsnap, thissnap);
1209 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1214 if (sdd->seento || !sdd->seenfrom) {
1219 istosnap = (strcmp(sdd->tosnap, thissnap) == 0);
1221 sdd->seento = B_TRUE;
1223 if (!sdd->doall && !isfromsnap && !istosnap) {
1224 if (sdd->replicate) {
1226 nvlist_t *snapprops;
1228 * Filter out all intermediate snapshots except origin
1229 * snapshots needed to replicate clones.
1231 nvlist_t *nvfs = fsavl_find(sdd->fsavl,
1232 zhp->zfs_dmustats.dds_guid, &snapname);
1234 VERIFY(0 == nvlist_lookup_nvlist(nvfs,
1235 "snapprops", &snapprops));
1236 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1237 thissnap, &snapprops));
1238 exclude = !nvlist_exists(snapprops, "is_clone_origin");
1245 * If a filter function exists, call it to determine whether
1246 * this snapshot will be sent.
1248 if (exclude || (sdd->filter_cb != NULL &&
1249 sdd->filter_cb(zhp, sdd->filter_cb_arg) == B_FALSE)) {
1251 * This snapshot is filtered out. Don't send it, and don't
1252 * set prevsnap_obj, so it will be as if this snapshot didn't
1253 * exist, and the next accepted snapshot will be sent as
1254 * an incremental from the last accepted one, or as the
1255 * first (and full) snapshot in the case of a replication,
1256 * non-incremental send.
1262 gather_holds(zhp, sdd);
1263 fromorigin = sdd->prevsnap[0] == '\0' &&
1264 (sdd->fromorigin || sdd->replicate);
1268 (void) estimate_ioctl(zhp, sdd->prevsnap_obj,
1271 send_print_verbose(fout, zhp->zfs_name,
1272 sdd->prevsnap[0] ? sdd->prevsnap : NULL,
1273 size, sdd->parsable);
1279 * If progress reporting is requested, spawn a new thread to
1280 * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1282 if (sdd->progress) {
1284 pa.pa_fd = sdd->outfd;
1285 pa.pa_parsable = sdd->parsable;
1287 if ((err = pthread_create(&tid, NULL,
1288 send_progress_thread, &pa)) != 0) {
1294 enum lzc_send_flags flags = 0;
1295 if (sdd->large_block)
1296 flags |= LZC_SEND_FLAG_LARGE_BLOCK;
1297 if (sdd->embed_data)
1298 flags |= LZC_SEND_FLAG_EMBED_DATA;
1300 err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
1301 fromorigin, sdd->outfd, flags, sdd->debugnv);
1303 if (sdd->progress) {
1304 (void) pthread_cancel(tid);
1305 (void) pthread_join(tid, NULL);
1309 (void) strcpy(sdd->prevsnap, thissnap);
1310 sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
1316 dump_filesystem(zfs_handle_t *zhp, void *arg)
1319 send_dump_data_t *sdd = arg;
1320 boolean_t missingfrom = B_FALSE;
1321 zfs_cmd_t zc = { 0 };
1323 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1324 zhp->zfs_name, sdd->tosnap);
1325 if (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1326 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1327 "WARNING: could not send %s@%s: does not exist\n"),
1328 zhp->zfs_name, sdd->tosnap);
1333 if (sdd->replicate && sdd->fromsnap) {
1335 * If this fs does not have fromsnap, and we're doing
1336 * recursive, we need to send a full stream from the
1337 * beginning (or an incremental from the origin if this
1338 * is a clone). If we're doing non-recursive, then let
1339 * them get the error.
1341 (void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
1342 zhp->zfs_name, sdd->fromsnap);
1343 if (ioctl(zhp->zfs_hdl->libzfs_fd,
1344 ZFS_IOC_OBJSET_STATS, &zc) != 0) {
1345 missingfrom = B_TRUE;
1349 sdd->seenfrom = sdd->seento = sdd->prevsnap[0] = 0;
1350 sdd->prevsnap_obj = 0;
1351 if (sdd->fromsnap == NULL || missingfrom)
1352 sdd->seenfrom = B_TRUE;
1354 rv = zfs_iter_snapshots_sorted(zhp, dump_snapshot, arg);
1355 if (!sdd->seenfrom) {
1356 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1357 "WARNING: could not send %s@%s:\n"
1358 "incremental source (%s@%s) does not exist\n"),
1359 zhp->zfs_name, sdd->tosnap,
1360 zhp->zfs_name, sdd->fromsnap);
1362 } else if (!sdd->seento) {
1363 if (sdd->fromsnap) {
1364 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1365 "WARNING: could not send %s@%s:\n"
1366 "incremental source (%s@%s) "
1367 "is not earlier than it\n"),
1368 zhp->zfs_name, sdd->tosnap,
1369 zhp->zfs_name, sdd->fromsnap);
1371 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1373 "could not send %s@%s: does not exist\n"),
1374 zhp->zfs_name, sdd->tosnap);
1383 dump_filesystems(zfs_handle_t *rzhp, void *arg)
1385 send_dump_data_t *sdd = arg;
1387 boolean_t needagain, progress;
1389 if (!sdd->replicate)
1390 return (dump_filesystem(rzhp, sdd));
1392 /* Mark the clone origin snapshots. */
1393 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1394 fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1396 uint64_t origin_guid = 0;
1398 VERIFY(0 == nvpair_value_nvlist(fspair, &nvfs));
1399 (void) nvlist_lookup_uint64(nvfs, "origin", &origin_guid);
1400 if (origin_guid != 0) {
1402 nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1403 origin_guid, &snapname);
1404 if (origin_nv != NULL) {
1405 nvlist_t *snapprops;
1406 VERIFY(0 == nvlist_lookup_nvlist(origin_nv,
1407 "snapprops", &snapprops));
1408 VERIFY(0 == nvlist_lookup_nvlist(snapprops,
1409 snapname, &snapprops));
1410 VERIFY(0 == nvlist_add_boolean(
1411 snapprops, "is_clone_origin"));
1416 needagain = progress = B_FALSE;
1417 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1418 fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1419 nvlist_t *fslist, *parent_nv;
1423 uint64_t origin_guid = 0;
1424 uint64_t parent_guid = 0;
1426 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1427 if (nvlist_lookup_boolean(fslist, "sent") == 0)
1430 VERIFY(nvlist_lookup_string(fslist, "name", &fsname) == 0);
1431 (void) nvlist_lookup_uint64(fslist, "origin", &origin_guid);
1432 (void) nvlist_lookup_uint64(fslist, "parentfromsnap",
1435 if (parent_guid != 0) {
1436 parent_nv = fsavl_find(sdd->fsavl, parent_guid, NULL);
1437 if (!nvlist_exists(parent_nv, "sent")) {
1438 /* parent has not been sent; skip this one */
1444 if (origin_guid != 0) {
1445 nvlist_t *origin_nv = fsavl_find(sdd->fsavl,
1447 if (origin_nv != NULL &&
1448 !nvlist_exists(origin_nv, "sent")) {
1450 * origin has not been sent yet;
1458 zhp = zfs_open(rzhp->zfs_hdl, fsname, ZFS_TYPE_DATASET);
1461 err = dump_filesystem(zhp, sdd);
1462 VERIFY(nvlist_add_boolean(fslist, "sent") == 0);
1473 /* clean out the sent flags in case we reuse this fss */
1474 for (fspair = nvlist_next_nvpair(sdd->fss, NULL); fspair;
1475 fspair = nvlist_next_nvpair(sdd->fss, fspair)) {
1478 VERIFY(nvpair_value_nvlist(fspair, &fslist) == 0);
1479 (void) nvlist_remove_all(fslist, "sent");
1486 zfs_send_resume_token_to_nvlist(libzfs_handle_t *hdl, const char *token)
1488 unsigned int version;
1490 unsigned long long checksum, packed_len;
1493 * Decode token header, which is:
1494 * <token version>-<checksum of payload>-<uncompressed payload length>
1495 * Note that the only supported token version is 1.
1497 nread = sscanf(token, "%u-%llx-%llx-",
1498 &version, &checksum, &packed_len);
1500 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1501 "resume token is corrupt (invalid format)"));
1505 if (version != ZFS_SEND_RESUME_TOKEN_VERSION) {
1506 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1507 "resume token is corrupt (invalid version %u)"),
1512 /* convert hexadecimal representation to binary */
1513 token = strrchr(token, '-') + 1;
1514 int len = strlen(token) / 2;
1515 unsigned char *compressed = zfs_alloc(hdl, len);
1516 for (int i = 0; i < len; i++) {
1517 nread = sscanf(token + i * 2, "%2hhx", compressed + i);
1520 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1521 "resume token is corrupt "
1522 "(payload is not hex-encoded)"));
1527 /* verify checksum */
1529 fletcher_4_native(compressed, len, NULL, &cksum);
1530 if (cksum.zc_word[0] != checksum) {
1532 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1533 "resume token is corrupt (incorrect checksum)"));
1538 void *packed = zfs_alloc(hdl, packed_len);
1539 uLongf packed_len_long = packed_len;
1540 if (uncompress(packed, &packed_len_long, compressed, len) != Z_OK ||
1541 packed_len_long != packed_len) {
1544 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1545 "resume token is corrupt (decompression failed)"));
1551 int error = nvlist_unpack(packed, packed_len, &nv, KM_SLEEP);
1555 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1556 "resume token is corrupt (nvlist_unpack failed)"));
1563 zfs_send_resume(libzfs_handle_t *hdl, sendflags_t *flags, int outfd,
1564 const char *resume_token)
1568 char *fromname = NULL;
1569 uint64_t resumeobj, resumeoff, toguid, fromguid, bytes;
1572 char name[ZFS_MAX_DATASET_NAME_LEN];
1573 enum lzc_send_flags lzc_flags = 0;
1575 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1576 "cannot resume send"));
1578 nvlist_t *resume_nvl =
1579 zfs_send_resume_token_to_nvlist(hdl, resume_token);
1580 if (resume_nvl == NULL) {
1582 * zfs_error_aux has already been set by
1583 * zfs_send_resume_token_to_nvlist
1585 return (zfs_error(hdl, EZFS_FAULT, errbuf));
1587 if (flags->verbose) {
1588 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
1589 "resume token contents:\n"));
1590 nvlist_print(stderr, resume_nvl);
1593 if (nvlist_lookup_string(resume_nvl, "toname", &toname) != 0 ||
1594 nvlist_lookup_uint64(resume_nvl, "object", &resumeobj) != 0 ||
1595 nvlist_lookup_uint64(resume_nvl, "offset", &resumeoff) != 0 ||
1596 nvlist_lookup_uint64(resume_nvl, "bytes", &bytes) != 0 ||
1597 nvlist_lookup_uint64(resume_nvl, "toguid", &toguid) != 0) {
1598 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1599 "resume token is corrupt"));
1600 return (zfs_error(hdl, EZFS_FAULT, errbuf));
1603 (void) nvlist_lookup_uint64(resume_nvl, "fromguid", &fromguid);
1605 if (flags->embed_data || nvlist_exists(resume_nvl, "embedok"))
1606 lzc_flags |= LZC_SEND_FLAG_EMBED_DATA;
1608 if (guid_to_name(hdl, toname, toguid, B_FALSE, name) != 0) {
1609 if (zfs_dataset_exists(hdl, toname, ZFS_TYPE_DATASET)) {
1610 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1611 "'%s' is no longer the same snapshot used in "
1612 "the initial send"), toname);
1614 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1615 "'%s' used in the initial send no longer exists"),
1618 return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1620 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
1622 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1623 "unable to access '%s'"), name);
1624 return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1627 if (fromguid != 0) {
1628 if (guid_to_name(hdl, toname, fromguid, B_TRUE, name) != 0) {
1629 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1630 "incremental source %#llx no longer exists"),
1631 (longlong_t)fromguid);
1632 return (zfs_error(hdl, EZFS_BADPATH, errbuf));
1637 if (flags->verbose) {
1639 error = lzc_send_space(zhp->zfs_name, fromname, &size);
1641 size = MAX(0, (int64_t)(size - bytes));
1642 send_print_verbose(stderr, zhp->zfs_name, fromname,
1643 size, flags->parsable);
1646 if (!flags->dryrun) {
1647 progress_arg_t pa = { 0 };
1650 * If progress reporting is requested, spawn a new thread to
1651 * poll ZFS_IOC_SEND_PROGRESS at a regular interval.
1653 if (flags->progress) {
1656 pa.pa_parsable = flags->parsable;
1658 error = pthread_create(&tid, NULL,
1659 send_progress_thread, &pa);
1666 error = lzc_send_resume(zhp->zfs_name, fromname, outfd,
1667 lzc_flags, resumeobj, resumeoff);
1669 if (flags->progress) {
1670 (void) pthread_cancel(tid);
1671 (void) pthread_join(tid, NULL);
1675 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1676 "warning: cannot send '%s'"), zhp->zfs_name);
1698 zfs_error_aux(hdl, strerror(errno));
1699 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
1702 return (zfs_standard_error(hdl, errno, errbuf));
1713 * Generate a send stream for the dataset identified by the argument zhp.
1715 * The content of the send stream is the snapshot identified by
1716 * 'tosnap'. Incremental streams are requested in two ways:
1717 * - from the snapshot identified by "fromsnap" (if non-null) or
1718 * - from the origin of the dataset identified by zhp, which must
1719 * be a clone. In this case, "fromsnap" is null and "fromorigin"
1722 * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
1723 * uses a special header (with a hdrtype field of DMU_COMPOUNDSTREAM)
1724 * if "replicate" is set. If "doall" is set, dump all the intermediate
1725 * snapshots. The DMU_COMPOUNDSTREAM header is used in the "doall"
1726 * case too. If "props" is set, send properties.
1729 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
1730 sendflags_t *flags, int outfd, snapfilter_cb_t filter_func,
1731 void *cb_arg, nvlist_t **debugnvp)
1734 send_dump_data_t sdd = { 0 };
1736 nvlist_t *fss = NULL;
1737 avl_tree_t *fsavl = NULL;
1738 static uint64_t holdseq;
1742 dedup_arg_t dda = { 0 };
1743 int featureflags = 0;
1746 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1747 "cannot send '%s'"), zhp->zfs_name);
1749 if (fromsnap && fromsnap[0] == '\0') {
1750 zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
1751 "zero-length incremental source"));
1752 return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
1755 if (zhp->zfs_type == ZFS_TYPE_FILESYSTEM) {
1757 version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
1758 if (version >= ZPL_VERSION_SA) {
1759 featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
1763 if (flags->dedup && !flags->dryrun) {
1764 featureflags |= (DMU_BACKUP_FEATURE_DEDUP |
1765 DMU_BACKUP_FEATURE_DEDUPPROPS);
1766 if ((err = pipe(pipefd)) != 0) {
1767 zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1768 return (zfs_error(zhp->zfs_hdl, EZFS_PIPEFAILED,
1771 dda.outputfd = outfd;
1772 dda.inputfd = pipefd[1];
1773 dda.dedup_hdl = zhp->zfs_hdl;
1774 if ((err = pthread_create(&tid, NULL, cksummer, &dda)) != 0) {
1775 (void) close(pipefd[0]);
1776 (void) close(pipefd[1]);
1777 zfs_error_aux(zhp->zfs_hdl, strerror(errno));
1778 return (zfs_error(zhp->zfs_hdl,
1779 EZFS_THREADCREATEFAILED, errbuf));
1783 if (flags->replicate || flags->doall || flags->props) {
1784 dmu_replay_record_t drr = { 0 };
1785 char *packbuf = NULL;
1787 zio_cksum_t zc = { 0 };
1789 if (flags->replicate || flags->props) {
1792 VERIFY(0 == nvlist_alloc(&hdrnv, NV_UNIQUE_NAME, 0));
1794 VERIFY(0 == nvlist_add_string(hdrnv,
1795 "fromsnap", fromsnap));
1797 VERIFY(0 == nvlist_add_string(hdrnv, "tosnap", tosnap));
1798 if (!flags->replicate) {
1799 VERIFY(0 == nvlist_add_boolean(hdrnv,
1803 err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
1804 fromsnap, tosnap, flags->replicate, flags->verbose,
1808 VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
1809 err = nvlist_pack(hdrnv, &packbuf, &buflen,
1819 if (!flags->dryrun) {
1820 /* write first begin record */
1821 drr.drr_type = DRR_BEGIN;
1822 drr.drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
1823 DMU_SET_STREAM_HDRTYPE(drr.drr_u.drr_begin.
1824 drr_versioninfo, DMU_COMPOUNDSTREAM);
1825 DMU_SET_FEATUREFLAGS(drr.drr_u.drr_begin.
1826 drr_versioninfo, featureflags);
1827 (void) snprintf(drr.drr_u.drr_begin.drr_toname,
1828 sizeof (drr.drr_u.drr_begin.drr_toname),
1829 "%s@%s", zhp->zfs_name, tosnap);
1830 drr.drr_payloadlen = buflen;
1832 err = dump_record(&drr, packbuf, buflen, &zc, outfd);
1837 /* write end record */
1838 bzero(&drr, sizeof (drr));
1839 drr.drr_type = DRR_END;
1840 drr.drr_u.drr_end.drr_checksum = zc;
1841 err = write(outfd, &drr, sizeof (drr));
1851 /* dump each stream */
1852 sdd.fromsnap = fromsnap;
1853 sdd.tosnap = tosnap;
1855 sdd.outfd = pipefd[0];
1858 sdd.replicate = flags->replicate;
1859 sdd.doall = flags->doall;
1860 sdd.fromorigin = flags->fromorigin;
1863 sdd.verbose = flags->verbose;
1864 sdd.parsable = flags->parsable;
1865 sdd.progress = flags->progress;
1866 sdd.dryrun = flags->dryrun;
1867 sdd.large_block = flags->largeblock;
1868 sdd.embed_data = flags->embed_data;
1869 sdd.filter_cb = filter_func;
1870 sdd.filter_cb_arg = cb_arg;
1872 sdd.debugnv = *debugnvp;
1873 if (sdd.verbose && sdd.dryrun)
1874 sdd.std_out = B_TRUE;
1875 fout = sdd.std_out ? stdout : stderr;
1878 * Some flags require that we place user holds on the datasets that are
1879 * being sent so they don't get destroyed during the send. We can skip
1880 * this step if the pool is imported read-only since the datasets cannot
1883 if (!flags->dryrun && !zpool_get_prop_int(zfs_get_pool_handle(zhp),
1884 ZPOOL_PROP_READONLY, NULL) &&
1885 zfs_spa_version(zhp, &spa_version) == 0 &&
1886 spa_version >= SPA_VERSION_USERREFS &&
1887 (flags->doall || flags->replicate)) {
1889 (void) snprintf(sdd.holdtag, sizeof (sdd.holdtag),
1890 ".send-%d-%llu", getpid(), (u_longlong_t)holdseq);
1891 sdd.cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
1892 if (sdd.cleanup_fd < 0) {
1896 sdd.snapholds = fnvlist_alloc();
1898 sdd.cleanup_fd = -1;
1899 sdd.snapholds = NULL;
1901 if (flags->verbose || sdd.snapholds != NULL) {
1903 * Do a verbose no-op dry run to get all the verbose output
1904 * or to gather snapshot hold's before generating any data,
1905 * then do a non-verbose real run to generate the streams.
1907 sdd.dryrun = B_TRUE;
1908 err = dump_filesystems(zhp, &sdd);
1913 if (flags->verbose) {
1914 if (flags->parsable) {
1915 (void) fprintf(fout, "size\t%llu\n",
1916 (longlong_t)sdd.size);
1919 zfs_nicenum(sdd.size, buf, sizeof (buf));
1920 (void) fprintf(fout, dgettext(TEXT_DOMAIN,
1921 "total estimated size is %s\n"), buf);
1925 /* Ensure no snaps found is treated as an error. */
1931 /* Skip the second run if dryrun was requested. */
1935 if (sdd.snapholds != NULL) {
1936 err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds);
1940 fnvlist_free(sdd.snapholds);
1941 sdd.snapholds = NULL;
1944 sdd.dryrun = B_FALSE;
1945 sdd.verbose = B_FALSE;
1948 err = dump_filesystems(zhp, &sdd);
1949 fsavl_destroy(fsavl);
1952 /* Ensure no snaps found is treated as an error. */
1953 if (err == 0 && !sdd.seento)
1958 (void) pthread_cancel(tid);
1959 (void) close(pipefd[0]);
1960 (void) pthread_join(tid, NULL);
1963 if (sdd.cleanup_fd != -1) {
1964 VERIFY(0 == close(sdd.cleanup_fd));
1965 sdd.cleanup_fd = -1;
1968 if (!flags->dryrun && (flags->replicate || flags->doall ||
1971 * write final end record. NB: want to do this even if
1972 * there was some error, because it might not be totally
1975 dmu_replay_record_t drr = { 0 };
1976 drr.drr_type = DRR_END;
1977 if (write(outfd, &drr, sizeof (drr)) == -1) {
1978 return (zfs_standard_error(zhp->zfs_hdl,
1983 return (err || sdd.err);
1986 err = zfs_standard_error(zhp->zfs_hdl, err, errbuf);
1988 fsavl_destroy(fsavl);
1990 fnvlist_free(sdd.snapholds);
1992 if (sdd.cleanup_fd != -1)
1993 VERIFY(0 == close(sdd.cleanup_fd));
1995 (void) pthread_cancel(tid);
1996 (void) close(pipefd[0]);
1997 (void) pthread_join(tid, NULL);
2003 zfs_send_one(zfs_handle_t *zhp, const char *from, int fd,
2004 enum lzc_send_flags flags)
2007 libzfs_handle_t *hdl = zhp->zfs_hdl;
2010 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2011 "warning: cannot send '%s'"), zhp->zfs_name);
2013 err = lzc_send(zhp->zfs_name, from, fd, flags);
2017 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2018 "not an earlier snapshot from the same fs"));
2019 return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
2023 if (lzc_exists(zhp->zfs_name)) {
2024 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2025 "incremental source (%s) does not exist"),
2028 return (zfs_error(hdl, EZFS_NOENT, errbuf));
2031 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2032 "target is busy; if a filesystem, "
2033 "it must not be mounted"));
2034 return (zfs_error(hdl, EZFS_BUSY, errbuf));
2049 zfs_error_aux(hdl, strerror(errno));
2050 return (zfs_error(hdl, EZFS_BADBACKUP, errbuf));
2053 return (zfs_standard_error(hdl, errno, errbuf));
2060 * Routines specific to "zfs recv"
2064 recv_read(libzfs_handle_t *hdl, int fd, void *buf, int ilen,
2065 boolean_t byteswap, zio_cksum_t *zc)
2071 assert(ilen <= SPA_MAXBLOCKSIZE);
2074 rv = read(fd, cp, len);
2079 if (rv < 0 || len != 0) {
2080 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2081 "failed to read from stream"));
2082 return (zfs_error(hdl, EZFS_BADSTREAM, dgettext(TEXT_DOMAIN,
2083 "cannot receive")));
2088 fletcher_4_incremental_byteswap(buf, ilen, zc);
2090 fletcher_4_incremental_native(buf, ilen, zc);
2096 recv_read_nvlist(libzfs_handle_t *hdl, int fd, int len, nvlist_t **nvp,
2097 boolean_t byteswap, zio_cksum_t *zc)
2102 buf = zfs_alloc(hdl, len);
2106 err = recv_read(hdl, fd, buf, len, byteswap, zc);
2112 err = nvlist_unpack(buf, len, nvp, 0);
2115 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
2116 "stream (malformed nvlist)"));
2123 recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
2124 int baselen, char *newname, recvflags_t *flags)
2127 zfs_cmd_t zc = { 0 };
2129 prop_changelist_t *clp;
2132 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
2135 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
2136 flags->force ? MS_FORCE : 0);
2140 err = changelist_prefix(clp);
2144 zc.zc_objset_type = DMU_OST_ZFS;
2145 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
2148 (void) strcpy(newname, tryname);
2150 (void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
2152 if (flags->verbose) {
2153 (void) printf("attempting rename %s to %s\n",
2154 zc.zc_name, zc.zc_value);
2156 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
2158 changelist_rename(clp, name, tryname);
2163 if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) {
2166 (void) snprintf(newname, ZFS_MAX_DATASET_NAME_LEN,
2167 "%.*srecv-%u-%u", baselen, name, getpid(), seq);
2168 (void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
2170 if (flags->verbose) {
2171 (void) printf("failed - trying rename %s to %s\n",
2172 zc.zc_name, zc.zc_value);
2174 err = ioctl(hdl->libzfs_fd, ZFS_IOC_RENAME, &zc);
2176 changelist_rename(clp, name, newname);
2177 if (err && flags->verbose) {
2178 (void) printf("failed (%u) - "
2179 "will try again on next pass\n", errno);
2182 } else if (flags->verbose) {
2184 (void) printf("success\n");
2186 (void) printf("failed (%u)\n", errno);
2189 (void) changelist_postfix(clp);
2190 changelist_free(clp);
2196 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
2197 char *newname, recvflags_t *flags)
2199 zfs_cmd_t zc = { 0 };
2201 prop_changelist_t *clp;
2203 boolean_t defer = B_FALSE;
2206 zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
2209 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
2210 flags->force ? MS_FORCE : 0);
2211 if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
2212 zfs_spa_version(zhp, &spa_version) == 0 &&
2213 spa_version >= SPA_VERSION_USERREFS)
2218 err = changelist_prefix(clp);
2222 zc.zc_objset_type = DMU_OST_ZFS;
2223 zc.zc_defer_destroy = defer;
2224 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
2227 (void) printf("attempting destroy %s\n", zc.zc_name);
2228 err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
2231 (void) printf("success\n");
2232 changelist_remove(clp, zc.zc_name);
2235 (void) changelist_postfix(clp);
2236 changelist_free(clp);
2239 * Deferred destroy might destroy the snapshot or only mark it to be
2240 * destroyed later, and it returns success in either case.
2242 if (err != 0 || (defer && zfs_dataset_exists(hdl, name,
2243 ZFS_TYPE_SNAPSHOT))) {
2244 err = recv_rename(hdl, name, NULL, baselen, newname, flags);
2250 typedef struct guid_to_name_data {
2252 boolean_t bookmark_ok;
2255 } guid_to_name_data_t;
2258 guid_to_name_cb(zfs_handle_t *zhp, void *arg)
2260 guid_to_name_data_t *gtnd = arg;
2264 if (gtnd->skip != NULL &&
2265 (slash = strrchr(zhp->zfs_name, '/')) != NULL &&
2266 strcmp(slash + 1, gtnd->skip) == 0) {
2271 if (zfs_prop_get_int(zhp, ZFS_PROP_GUID) == gtnd->guid) {
2272 (void) strcpy(gtnd->name, zhp->zfs_name);
2277 err = zfs_iter_children(zhp, guid_to_name_cb, gtnd);
2278 if (err != EEXIST && gtnd->bookmark_ok)
2279 err = zfs_iter_bookmarks(zhp, guid_to_name_cb, gtnd);
2285 * Attempt to find the local dataset associated with this guid. In the case of
2286 * multiple matches, we attempt to find the "best" match by searching
2287 * progressively larger portions of the hierarchy. This allows one to send a
2288 * tree of datasets individually and guarantee that we will find the source
2289 * guid within that hierarchy, even if there are multiple matches elsewhere.
2292 guid_to_name(libzfs_handle_t *hdl, const char *parent, uint64_t guid,
2293 boolean_t bookmark_ok, char *name)
2295 char pname[ZFS_MAX_DATASET_NAME_LEN];
2296 guid_to_name_data_t gtnd;
2299 gtnd.bookmark_ok = bookmark_ok;
2304 * Search progressively larger portions of the hierarchy, starting
2305 * with the filesystem specified by 'parent'. This will
2306 * select the "most local" version of the origin snapshot in the case
2307 * that there are multiple matching snapshots in the system.
2309 (void) strlcpy(pname, parent, sizeof (pname));
2310 char *cp = strrchr(pname, '@');
2312 cp = strchr(pname, '\0');
2313 for (; cp != NULL; cp = strrchr(pname, '/')) {
2314 /* Chop off the last component and open the parent */
2316 zfs_handle_t *zhp = make_dataset_handle(hdl, pname);
2320 int err = guid_to_name_cb(zfs_handle_dup(zhp), >nd);
2322 err = zfs_iter_children(zhp, guid_to_name_cb, >nd);
2323 if (err != EEXIST && bookmark_ok)
2324 err = zfs_iter_bookmarks(zhp, guid_to_name_cb, >nd);
2330 * Remember the last portion of the dataset so we skip it next
2331 * time through (as we've already searched that portion of the
2334 gtnd.skip = strrchr(pname, '/') + 1;
2341 * Return +1 if guid1 is before guid2, 0 if they are the same, and -1 if
2342 * guid1 is after guid2.
2345 created_before(libzfs_handle_t *hdl, avl_tree_t *avl,
2346 uint64_t guid1, uint64_t guid2)
2349 char *fsname, *snapname;
2350 char buf[ZFS_MAX_DATASET_NAME_LEN];
2352 zfs_handle_t *guid1hdl, *guid2hdl;
2353 uint64_t create1, create2;
2360 nvfs = fsavl_find(avl, guid1, &snapname);
2361 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2362 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
2363 guid1hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
2364 if (guid1hdl == NULL)
2367 nvfs = fsavl_find(avl, guid2, &snapname);
2368 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2369 (void) snprintf(buf, sizeof (buf), "%s@%s", fsname, snapname);
2370 guid2hdl = zfs_open(hdl, buf, ZFS_TYPE_SNAPSHOT);
2371 if (guid2hdl == NULL) {
2372 zfs_close(guid1hdl);
2376 create1 = zfs_prop_get_int(guid1hdl, ZFS_PROP_CREATETXG);
2377 create2 = zfs_prop_get_int(guid2hdl, ZFS_PROP_CREATETXG);
2379 if (create1 < create2)
2381 else if (create1 > create2)
2386 zfs_close(guid1hdl);
2387 zfs_close(guid2hdl);
2393 recv_incremental_replication(libzfs_handle_t *hdl, const char *tofs,
2394 recvflags_t *flags, nvlist_t *stream_nv, avl_tree_t *stream_avl,
2397 nvlist_t *local_nv, *deleted = NULL;
2398 avl_tree_t *local_avl;
2399 nvpair_t *fselem, *nextfselem;
2401 char newname[ZFS_MAX_DATASET_NAME_LEN];
2404 boolean_t needagain, progress, recursive;
2407 VERIFY(0 == nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap));
2409 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2416 needagain = progress = B_FALSE;
2418 VERIFY(0 == nvlist_alloc(&deleted, NV_UNIQUE_NAME, 0));
2420 if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
2421 recursive, B_FALSE, &local_nv, &local_avl)) != 0)
2425 * Process deletes and renames
2427 for (fselem = nvlist_next_nvpair(local_nv, NULL);
2428 fselem; fselem = nextfselem) {
2429 nvlist_t *nvfs, *snaps;
2430 nvlist_t *stream_nvfs = NULL;
2431 nvpair_t *snapelem, *nextsnapelem;
2432 uint64_t fromguid = 0;
2433 uint64_t originguid = 0;
2434 uint64_t stream_originguid = 0;
2435 uint64_t parent_fromsnap_guid, stream_parent_fromsnap_guid;
2436 char *fsname, *stream_fsname;
2438 nextfselem = nvlist_next_nvpair(local_nv, fselem);
2440 VERIFY(0 == nvpair_value_nvlist(fselem, &nvfs));
2441 VERIFY(0 == nvlist_lookup_nvlist(nvfs, "snaps", &snaps));
2442 VERIFY(0 == nvlist_lookup_string(nvfs, "name", &fsname));
2443 VERIFY(0 == nvlist_lookup_uint64(nvfs, "parentfromsnap",
2444 &parent_fromsnap_guid));
2445 (void) nvlist_lookup_uint64(nvfs, "origin", &originguid);
2448 * First find the stream's fs, so we can check for
2449 * a different origin (due to "zfs promote")
2451 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2452 snapelem; snapelem = nvlist_next_nvpair(snaps, snapelem)) {
2455 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2456 stream_nvfs = fsavl_find(stream_avl, thisguid, NULL);
2458 if (stream_nvfs != NULL)
2462 /* check for promote */
2463 (void) nvlist_lookup_uint64(stream_nvfs, "origin",
2464 &stream_originguid);
2465 if (stream_nvfs && originguid != stream_originguid) {
2466 switch (created_before(hdl, local_avl,
2467 stream_originguid, originguid)) {
2470 zfs_cmd_t zc = { 0 };
2471 nvlist_t *origin_nvfs;
2472 char *origin_fsname;
2475 (void) printf("promoting %s\n", fsname);
2477 origin_nvfs = fsavl_find(local_avl, originguid,
2479 VERIFY(0 == nvlist_lookup_string(origin_nvfs,
2480 "name", &origin_fsname));
2481 (void) strlcpy(zc.zc_value, origin_fsname,
2482 sizeof (zc.zc_value));
2483 (void) strlcpy(zc.zc_name, fsname,
2484 sizeof (zc.zc_name));
2485 error = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
2493 fsavl_destroy(local_avl);
2494 nvlist_free(local_nv);
2498 * We had/have the wrong origin, therefore our
2499 * list of snapshots is wrong. Need to handle
2500 * them on the next pass.
2506 for (snapelem = nvlist_next_nvpair(snaps, NULL);
2507 snapelem; snapelem = nextsnapelem) {
2509 char *stream_snapname;
2510 nvlist_t *found, *props;
2512 nextsnapelem = nvlist_next_nvpair(snaps, snapelem);
2514 VERIFY(0 == nvpair_value_uint64(snapelem, &thisguid));
2515 found = fsavl_find(stream_avl, thisguid,
2518 /* check for delete */
2519 if (found == NULL) {
2520 char name[ZFS_MAX_DATASET_NAME_LEN];
2525 (void) snprintf(name, sizeof (name), "%s@%s",
2526 fsname, nvpair_name(snapelem));
2528 error = recv_destroy(hdl, name,
2529 strlen(fsname)+1, newname, flags);
2534 sprintf(guidname, "%lu", thisguid);
2535 nvlist_add_boolean(deleted, guidname);
2539 stream_nvfs = found;
2541 if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
2542 &props) && 0 == nvlist_lookup_nvlist(props,
2543 stream_snapname, &props)) {
2544 zfs_cmd_t zc = { 0 };
2546 zc.zc_cookie = B_TRUE; /* received */
2547 (void) snprintf(zc.zc_name, sizeof (zc.zc_name),
2548 "%s@%s", fsname, nvpair_name(snapelem));
2549 if (zcmd_write_src_nvlist(hdl, &zc,
2551 (void) zfs_ioctl(hdl,
2552 ZFS_IOC_SET_PROP, &zc);
2553 zcmd_free_nvlists(&zc);
2557 /* check for different snapname */
2558 if (strcmp(nvpair_name(snapelem),
2559 stream_snapname) != 0) {
2560 char name[ZFS_MAX_DATASET_NAME_LEN];
2561 char tryname[ZFS_MAX_DATASET_NAME_LEN];
2563 (void) snprintf(name, sizeof (name), "%s@%s",
2564 fsname, nvpair_name(snapelem));
2565 (void) snprintf(tryname, sizeof (name), "%s@%s",
2566 fsname, stream_snapname);
2568 error = recv_rename(hdl, name, tryname,
2569 strlen(fsname)+1, newname, flags);
2576 if (strcmp(stream_snapname, fromsnap) == 0)
2577 fromguid = thisguid;
2580 /* check for delete */
2581 if (stream_nvfs == NULL) {
2585 error = recv_destroy(hdl, fsname, strlen(tofs)+1,
2591 sprintf(guidname, "%lu", parent_fromsnap_guid);
2592 nvlist_add_boolean(deleted, guidname);
2596 if (fromguid == 0) {
2597 if (flags->verbose) {
2598 (void) printf("local fs %s does not have "
2599 "fromsnap (%s in stream); must have "
2600 "been deleted locally; ignoring\n",
2606 VERIFY(0 == nvlist_lookup_string(stream_nvfs,
2607 "name", &stream_fsname));
2608 VERIFY(0 == nvlist_lookup_uint64(stream_nvfs,
2609 "parentfromsnap", &stream_parent_fromsnap_guid));
2611 s1 = strrchr(fsname, '/');
2612 s2 = strrchr(stream_fsname, '/');
2615 * Check if we're going to rename based on parent guid change
2616 * and the current parent guid was also deleted. If it was then
2617 * rename will fail and is likely unneeded, so avoid this and
2618 * force an early retry to determine the new
2619 * parent_fromsnap_guid.
2621 if (stream_parent_fromsnap_guid != 0 &&
2622 parent_fromsnap_guid != 0 &&
2623 stream_parent_fromsnap_guid != parent_fromsnap_guid) {
2624 sprintf(guidname, "%lu", parent_fromsnap_guid);
2625 if (nvlist_exists(deleted, guidname)) {
2633 * Check for rename. If the exact receive path is specified, it
2634 * does not count as a rename, but we still need to check the
2635 * datasets beneath it.
2637 if ((stream_parent_fromsnap_guid != 0 &&
2638 parent_fromsnap_guid != 0 &&
2639 stream_parent_fromsnap_guid != parent_fromsnap_guid) ||
2640 ((flags->isprefix || strcmp(tofs, fsname) != 0) &&
2641 (s1 != NULL) && (s2 != NULL) && strcmp(s1, s2) != 0)) {
2643 char tryname[ZFS_MAX_DATASET_NAME_LEN];
2645 parent = fsavl_find(local_avl,
2646 stream_parent_fromsnap_guid, NULL);
2648 * NB: parent might not be found if we used the
2649 * tosnap for stream_parent_fromsnap_guid,
2650 * because the parent is a newly-created fs;
2651 * we'll be able to rename it after we recv the
2654 if (parent != NULL) {
2657 VERIFY(0 == nvlist_lookup_string(parent, "name",
2659 (void) snprintf(tryname, sizeof (tryname),
2660 "%s%s", pname, strrchr(stream_fsname, '/'));
2663 if (flags->verbose) {
2664 (void) printf("local fs %s new parent "
2665 "not found\n", fsname);
2671 error = recv_rename(hdl, fsname, tryname,
2672 strlen(tofs)+1, newname, flags);
2674 if (renamed != NULL && newname[0] != '\0') {
2675 VERIFY(0 == nvlist_add_boolean(renamed,
2687 fsavl_destroy(local_avl);
2688 nvlist_free(local_nv);
2689 nvlist_free(deleted);
2691 if (needagain && progress) {
2692 /* do another pass to fix up temporary names */
2694 (void) printf("another pass:\n");
2702 zfs_receive_package(libzfs_handle_t *hdl, int fd, const char *destname,
2703 recvflags_t *flags, dmu_replay_record_t *drr, zio_cksum_t *zc,
2704 char **top_zfs, int cleanup_fd, uint64_t *action_handlep)
2706 nvlist_t *stream_nv = NULL;
2707 avl_tree_t *stream_avl = NULL;
2708 char *fromsnap = NULL;
2709 char *sendsnap = NULL;
2711 char tofs[ZFS_MAX_DATASET_NAME_LEN];
2712 char sendfs[ZFS_MAX_DATASET_NAME_LEN];
2714 dmu_replay_record_t drre;
2716 boolean_t anyerr = B_FALSE;
2717 boolean_t softerr = B_FALSE;
2718 boolean_t recursive;
2720 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2723 assert(drr->drr_type == DRR_BEGIN);
2724 assert(drr->drr_u.drr_begin.drr_magic == DMU_BACKUP_MAGIC);
2725 assert(DMU_GET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo) ==
2726 DMU_COMPOUNDSTREAM);
2729 * Read in the nvlist from the stream.
2731 if (drr->drr_payloadlen != 0) {
2732 error = recv_read_nvlist(hdl, fd, drr->drr_payloadlen,
2733 &stream_nv, flags->byteswap, zc);
2735 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2740 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
2743 if (recursive && strchr(destname, '@')) {
2744 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2745 "cannot specify snapshot name for multi-snapshot stream"));
2746 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2751 * Read in the end record and verify checksum.
2753 if (0 != (error = recv_read(hdl, fd, &drre, sizeof (drre),
2754 flags->byteswap, NULL)))
2756 if (flags->byteswap) {
2757 drre.drr_type = BSWAP_32(drre.drr_type);
2758 drre.drr_u.drr_end.drr_checksum.zc_word[0] =
2759 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[0]);
2760 drre.drr_u.drr_end.drr_checksum.zc_word[1] =
2761 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[1]);
2762 drre.drr_u.drr_end.drr_checksum.zc_word[2] =
2763 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[2]);
2764 drre.drr_u.drr_end.drr_checksum.zc_word[3] =
2765 BSWAP_64(drre.drr_u.drr_end.drr_checksum.zc_word[3]);
2767 if (drre.drr_type != DRR_END) {
2768 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2771 if (!ZIO_CHECKSUM_EQUAL(drre.drr_u.drr_end.drr_checksum, *zc)) {
2772 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2773 "incorrect header checksum"));
2774 error = zfs_error(hdl, EZFS_BADSTREAM, errbuf);
2778 (void) nvlist_lookup_string(stream_nv, "fromsnap", &fromsnap);
2780 if (drr->drr_payloadlen != 0) {
2781 nvlist_t *stream_fss;
2783 VERIFY(0 == nvlist_lookup_nvlist(stream_nv, "fss",
2785 if ((stream_avl = fsavl_create(stream_fss)) == NULL) {
2786 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2787 "couldn't allocate avl tree"));
2788 error = zfs_error(hdl, EZFS_NOMEM, errbuf);
2792 if (fromsnap != NULL) {
2793 nvlist_t *renamed = NULL;
2794 nvpair_t *pair = NULL;
2796 (void) strlcpy(tofs, destname, sizeof (tofs));
2797 if (flags->isprefix) {
2798 struct drr_begin *drrb = &drr->drr_u.drr_begin;
2801 if (flags->istail) {
2802 cp = strrchr(drrb->drr_toname, '/');
2804 (void) strlcat(tofs, "/",
2808 i = (cp - drrb->drr_toname);
2811 i = strcspn(drrb->drr_toname, "/@");
2813 /* zfs_receive_one() will create_parents() */
2814 (void) strlcat(tofs, &drrb->drr_toname[i],
2816 *strchr(tofs, '@') = '\0';
2819 if (recursive && !flags->dryrun && !flags->nomount) {
2820 VERIFY(0 == nvlist_alloc(&renamed,
2821 NV_UNIQUE_NAME, 0));
2824 softerr = recv_incremental_replication(hdl, tofs, flags,
2825 stream_nv, stream_avl, renamed);
2827 /* Unmount renamed filesystems before receiving. */
2828 while ((pair = nvlist_next_nvpair(renamed,
2831 prop_changelist_t *clp = NULL;
2833 zhp = zfs_open(hdl, nvpair_name(pair),
2834 ZFS_TYPE_FILESYSTEM);
2836 clp = changelist_gather(zhp,
2837 ZFS_PROP_MOUNTPOINT, 0, 0);
2841 changelist_prefix(clp);
2842 changelist_free(clp);
2847 nvlist_free(renamed);
2852 * Get the fs specified by the first path in the stream (the top level
2853 * specified by 'zfs send') and pass it to each invocation of
2854 * zfs_receive_one().
2856 (void) strlcpy(sendfs, drr->drr_u.drr_begin.drr_toname,
2858 if ((cp = strchr(sendfs, '@')) != NULL) {
2861 * Find the "sendsnap", the final snapshot in a replication
2862 * stream. zfs_receive_one() handles certain errors
2863 * differently, depending on if the contained stream is the
2866 sendsnap = (cp + 1);
2869 /* Finally, receive each contained stream */
2872 * we should figure out if it has a recoverable
2873 * error, in which case do a recv_skip() and drive on.
2874 * Note, if we fail due to already having this guid,
2875 * zfs_receive_one() will take care of it (ie,
2876 * recv_skip() and return 0).
2878 error = zfs_receive_impl(hdl, destname, NULL, flags, fd,
2879 sendfs, stream_nv, stream_avl, top_zfs, cleanup_fd,
2880 action_handlep, sendsnap);
2881 if (error == ENODATA) {
2886 } while (error == 0);
2888 if (drr->drr_payloadlen != 0 && fromsnap != NULL) {
2890 * Now that we have the fs's they sent us, try the
2893 softerr = recv_incremental_replication(hdl, tofs, flags,
2894 stream_nv, stream_avl, NULL);
2898 fsavl_destroy(stream_avl);
2899 nvlist_free(stream_nv);
2908 trunc_prop_errs(int truncated)
2910 ASSERT(truncated != 0);
2913 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2914 "1 more property could not be set\n"));
2916 (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
2917 "%d more properties could not be set\n"), truncated);
2921 recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
2923 dmu_replay_record_t *drr;
2924 void *buf = zfs_alloc(hdl, SPA_MAXBLOCKSIZE);
2927 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2928 "cannot receive:"));
2930 /* XXX would be great to use lseek if possible... */
2933 while (recv_read(hdl, fd, drr, sizeof (dmu_replay_record_t),
2934 byteswap, NULL) == 0) {
2936 drr->drr_type = BSWAP_32(drr->drr_type);
2938 switch (drr->drr_type) {
2940 if (drr->drr_payloadlen != 0) {
2941 (void) recv_read(hdl, fd, buf,
2942 drr->drr_payloadlen, B_FALSE, NULL);
2952 drr->drr_u.drr_object.drr_bonuslen =
2953 BSWAP_32(drr->drr_u.drr_object.
2956 (void) recv_read(hdl, fd, buf,
2957 P2ROUNDUP(drr->drr_u.drr_object.drr_bonuslen, 8),
2963 drr->drr_u.drr_write.drr_length =
2964 BSWAP_64(drr->drr_u.drr_write.drr_length);
2966 (void) recv_read(hdl, fd, buf,
2967 drr->drr_u.drr_write.drr_length, B_FALSE, NULL);
2971 drr->drr_u.drr_spill.drr_length =
2972 BSWAP_64(drr->drr_u.drr_spill.drr_length);
2974 (void) recv_read(hdl, fd, buf,
2975 drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
2977 case DRR_WRITE_EMBEDDED:
2979 drr->drr_u.drr_write_embedded.drr_psize =
2980 BSWAP_32(drr->drr_u.drr_write_embedded.
2983 (void) recv_read(hdl, fd, buf,
2984 P2ROUNDUP(drr->drr_u.drr_write_embedded.drr_psize,
2987 case DRR_WRITE_BYREF:
2988 case DRR_FREEOBJECTS:
2993 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2994 "invalid record type"));
2995 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3004 recv_ecksum_set_aux(libzfs_handle_t *hdl, const char *target_snap,
3005 boolean_t resumable)
3007 char target_fs[ZFS_MAX_DATASET_NAME_LEN];
3009 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3010 "checksum mismatch or incomplete stream"));
3014 (void) strlcpy(target_fs, target_snap, sizeof (target_fs));
3015 *strchr(target_fs, '@') = '\0';
3016 zfs_handle_t *zhp = zfs_open(hdl, target_fs,
3017 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
3021 char token_buf[ZFS_MAXPROPLEN];
3022 int error = zfs_prop_get(zhp, ZFS_PROP_RECEIVE_RESUME_TOKEN,
3023 token_buf, sizeof (token_buf),
3024 NULL, NULL, 0, B_TRUE);
3026 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3027 "checksum mismatch or incomplete stream.\n"
3028 "Partially received snapshot is saved.\n"
3029 "A resuming stream can be generated on the sending "
3030 "system by running:\n"
3038 * Restores a backup of tosnap from the file descriptor specified by infd.
3041 zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
3042 const char *originsnap, recvflags_t *flags, dmu_replay_record_t *drr,
3043 dmu_replay_record_t *drr_noswap, const char *sendfs, nvlist_t *stream_nv,
3044 avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
3045 uint64_t *action_handlep, const char *finalsnap)
3047 zfs_cmd_t zc = { 0 };
3049 int ioctl_err, ioctl_errno, err;
3051 struct drr_begin *drrb = &drr->drr_u.drr_begin;
3053 char prop_errbuf[1024];
3054 const char *chopprefix;
3055 boolean_t newfs = B_FALSE;
3056 boolean_t stream_wantsnewfs;
3057 uint64_t parent_snapguid = 0;
3058 prop_changelist_t *clp = NULL;
3059 nvlist_t *snapprops_nvlist = NULL;
3060 zprop_errflags_t prop_errflags;
3061 boolean_t recursive;
3062 char *snapname = NULL;
3064 begin_time = time(NULL);
3066 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3069 recursive = (nvlist_lookup_boolean(stream_nv, "not_recursive") ==
3072 if (stream_avl != NULL) {
3073 nvlist_t *fs = fsavl_find(stream_avl, drrb->drr_toguid,
3078 (void) nvlist_lookup_uint64(fs, "parentfromsnap",
3080 err = nvlist_lookup_nvlist(fs, "props", &props);
3082 VERIFY(0 == nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
3084 if (flags->canmountoff) {
3085 VERIFY(0 == nvlist_add_uint64(props,
3086 zfs_prop_to_name(ZFS_PROP_CANMOUNT), 0));
3088 ret = zcmd_write_src_nvlist(hdl, &zc, props);
3092 if (0 == nvlist_lookup_nvlist(fs, "snapprops", &props)) {
3093 VERIFY(0 == nvlist_lookup_nvlist(props,
3094 snapname, &snapprops_nvlist));
3104 * Determine how much of the snapshot name stored in the stream
3105 * we are going to tack on to the name they specified on the
3106 * command line, and how much we are going to chop off.
3108 * If they specified a snapshot, chop the entire name stored in
3111 if (flags->istail) {
3113 * A filesystem was specified with -e. We want to tack on only
3114 * the tail of the sent snapshot path.
3116 if (strchr(tosnap, '@')) {
3117 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3118 "argument - snapshot not allowed with -e"));
3119 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3122 chopprefix = strrchr(sendfs, '/');
3124 if (chopprefix == NULL) {
3126 * The tail is the poolname, so we need to
3127 * prepend a path separator.
3129 int len = strlen(drrb->drr_toname);
3130 cp = malloc(len + 2);
3132 (void) strcpy(&cp[1], drrb->drr_toname);
3135 chopprefix = drrb->drr_toname + (chopprefix - sendfs);
3137 } else if (flags->isprefix) {
3139 * A filesystem was specified with -d. We want to tack on
3140 * everything but the first element of the sent snapshot path
3141 * (all but the pool name).
3143 if (strchr(tosnap, '@')) {
3144 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3145 "argument - snapshot not allowed with -d"));
3146 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3149 chopprefix = strchr(drrb->drr_toname, '/');
3150 if (chopprefix == NULL)
3151 chopprefix = strchr(drrb->drr_toname, '@');
3152 } else if (strchr(tosnap, '@') == NULL) {
3154 * If a filesystem was specified without -d or -e, we want to
3155 * tack on everything after the fs specified by 'zfs send'.
3157 chopprefix = drrb->drr_toname + strlen(sendfs);
3159 /* A snapshot was specified as an exact path (no -d or -e). */
3161 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3162 "cannot specify snapshot name for multi-snapshot "
3164 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3166 chopprefix = drrb->drr_toname + strlen(drrb->drr_toname);
3169 ASSERT(strstr(drrb->drr_toname, sendfs) == drrb->drr_toname);
3170 ASSERT(chopprefix > drrb->drr_toname);
3171 ASSERT(chopprefix <= drrb->drr_toname + strlen(drrb->drr_toname));
3172 ASSERT(chopprefix[0] == '/' || chopprefix[0] == '@' ||
3173 chopprefix[0] == '\0');
3176 * Determine name of destination snapshot, store in zc_value.
3178 (void) strcpy(zc.zc_value, tosnap);
3179 (void) strncat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
3181 if (zfs_ioctl_version == ZFS_IOCVER_UNDEF)
3182 zfs_ioctl_version = get_zfs_ioctl_version();
3184 * For forward compatibility hide tosnap in zc_value
3186 if (zfs_ioctl_version < ZFS_IOCVER_LZC)
3187 (void) strcpy(zc.zc_value + strlen(zc.zc_value) + 1, tosnap);
3190 if (!zfs_name_valid(zc.zc_value, ZFS_TYPE_SNAPSHOT)) {
3191 zcmd_free_nvlists(&zc);
3192 return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3196 * Determine the name of the origin snapshot, store in zc_string.
3199 (void) strncpy(zc.zc_string, originsnap, sizeof (zc.zc_string));
3201 (void) printf("using provided clone origin %s\n",
3203 } else if (drrb->drr_flags & DRR_FLAG_CLONE) {
3204 if (guid_to_name(hdl, zc.zc_value,
3205 drrb->drr_fromguid, B_FALSE, zc.zc_string) != 0) {
3206 zcmd_free_nvlists(&zc);
3207 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3208 "local origin for clone %s does not exist"),
3210 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3213 (void) printf("found clone origin %s\n", zc.zc_string);
3216 boolean_t resuming = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
3217 DMU_BACKUP_FEATURE_RESUMING;
3218 stream_wantsnewfs = (drrb->drr_fromguid == 0 ||
3219 (drrb->drr_flags & DRR_FLAG_CLONE) || originsnap) && !resuming;
3221 if (stream_wantsnewfs) {
3223 * if the parent fs does not exist, look for it based on
3224 * the parent snap GUID
3226 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3227 "cannot receive new filesystem stream"));
3229 (void) strcpy(zc.zc_name, zc.zc_value);
3230 cp = strrchr(zc.zc_name, '/');
3234 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
3235 char suffix[ZFS_MAX_DATASET_NAME_LEN];
3236 (void) strcpy(suffix, strrchr(zc.zc_value, '/'));
3237 if (guid_to_name(hdl, zc.zc_name, parent_snapguid,
3238 B_FALSE, zc.zc_value) == 0) {
3239 *strchr(zc.zc_value, '@') = '\0';
3240 (void) strcat(zc.zc_value, suffix);
3245 * if the fs does not exist, look for it based on the
3248 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3249 "cannot receive incremental stream"));
3251 (void) strcpy(zc.zc_name, zc.zc_value);
3252 *strchr(zc.zc_name, '@') = '\0';
3255 * If the exact receive path was specified and this is the
3256 * topmost path in the stream, then if the fs does not exist we
3257 * should look no further.
3259 if ((flags->isprefix || (*(chopprefix = drrb->drr_toname +
3260 strlen(sendfs)) != '\0' && *chopprefix != '@')) &&
3261 !zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
3262 char snap[ZFS_MAX_DATASET_NAME_LEN];
3263 (void) strcpy(snap, strchr(zc.zc_value, '@'));
3264 if (guid_to_name(hdl, zc.zc_name, drrb->drr_fromguid,
3265 B_FALSE, zc.zc_value) == 0) {
3266 *strchr(zc.zc_value, '@') = '\0';
3267 (void) strcat(zc.zc_value, snap);
3272 (void) strcpy(zc.zc_name, zc.zc_value);
3273 *strchr(zc.zc_name, '@') = '\0';
3275 if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
3279 * Destination fs exists. It must be one of these cases:
3280 * - an incremental send stream
3281 * - the stream specifies a new fs (full stream or clone)
3282 * and they want us to blow away the existing fs (and
3283 * have therefore specified -F and removed any snapshots)
3284 * - we are resuming a failed receive.
3286 if (stream_wantsnewfs) {
3287 if (!flags->force) {
3288 zcmd_free_nvlists(&zc);
3289 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3290 "destination '%s' exists\n"
3291 "must specify -F to overwrite it"),
3293 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3295 if (ioctl(hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
3297 zcmd_free_nvlists(&zc);
3298 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3299 "destination has snapshots (eg. %s)\n"
3300 "must destroy them to overwrite it"),
3302 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3306 if ((zhp = zfs_open(hdl, zc.zc_name,
3307 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
3308 zcmd_free_nvlists(&zc);
3312 if (stream_wantsnewfs &&
3313 zhp->zfs_dmustats.dds_origin[0]) {
3314 zcmd_free_nvlists(&zc);
3316 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3317 "destination '%s' is a clone\n"
3318 "must destroy it to overwrite it"),
3320 return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3323 if (!flags->dryrun && zhp->zfs_type == ZFS_TYPE_FILESYSTEM &&
3324 stream_wantsnewfs) {
3325 /* We can't do online recv in this case */
3326 clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
3329 zcmd_free_nvlists(&zc);
3332 if (changelist_prefix(clp) != 0) {
3333 changelist_free(clp);
3335 zcmd_free_nvlists(&zc);
3341 * If we are resuming a newfs, set newfs here so that we will
3342 * mount it if the recv succeeds this time. We can tell
3343 * that it was a newfs on the first recv because the fs
3344 * itself will be inconsistent (if the fs existed when we
3345 * did the first recv, we would have received it into
3348 if (resuming && zfs_prop_get_int(zhp, ZFS_PROP_INCONSISTENT))
3354 * Destination filesystem does not exist. Therefore we better
3355 * be creating a new filesystem (either from a full backup, or
3356 * a clone). It would therefore be invalid if the user
3357 * specified only the pool name (i.e. if the destination name
3358 * contained no slash character).
3360 if (!stream_wantsnewfs ||
3361 (cp = strrchr(zc.zc_name, '/')) == NULL) {
3362 zcmd_free_nvlists(&zc);
3363 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3364 "destination '%s' does not exist"), zc.zc_name);
3365 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3369 * Trim off the final dataset component so we perform the
3370 * recvbackup ioctl to the filesystems's parent.
3374 if (flags->isprefix && !flags->istail && !flags->dryrun &&
3375 create_parents(hdl, zc.zc_value, strlen(tosnap)) != 0) {
3376 zcmd_free_nvlists(&zc);
3377 return (zfs_error(hdl, EZFS_BADRESTORE, errbuf));
3383 zc.zc_begin_record = *drr_noswap;
3384 zc.zc_cookie = infd;
3385 zc.zc_guid = flags->force;
3386 zc.zc_resumable = flags->resumable;
3387 if (flags->verbose) {
3388 (void) printf("%s %s stream of %s into %s\n",
3389 flags->dryrun ? "would receive" : "receiving",
3390 drrb->drr_fromguid ? "incremental" : "full",
3391 drrb->drr_toname, zc.zc_value);
3392 (void) fflush(stdout);
3395 if (flags->dryrun) {
3396 zcmd_free_nvlists(&zc);
3397 return (recv_skip(hdl, infd, flags->byteswap));
3400 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)prop_errbuf;
3401 zc.zc_nvlist_dst_size = sizeof (prop_errbuf);
3402 zc.zc_cleanup_fd = cleanup_fd;
3403 zc.zc_action_handle = *action_handlep;
3405 err = ioctl_err = zfs_ioctl(hdl, ZFS_IOC_RECV, &zc);
3406 ioctl_errno = errno;
3407 prop_errflags = (zprop_errflags_t)zc.zc_obj;
3410 nvlist_t *prop_errors;
3411 VERIFY(0 == nvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
3412 zc.zc_nvlist_dst_size, &prop_errors, 0));
3414 nvpair_t *prop_err = NULL;
3416 while ((prop_err = nvlist_next_nvpair(prop_errors,
3417 prop_err)) != NULL) {
3422 prop = zfs_name_to_prop(nvpair_name(prop_err));
3423 (void) nvpair_value_int32(prop_err, &intval);
3424 if (strcmp(nvpair_name(prop_err),
3425 ZPROP_N_MORE_ERRORS) == 0) {
3426 trunc_prop_errs(intval);
3428 } else if (snapname == NULL || finalsnap == NULL ||
3429 strcmp(finalsnap, snapname) == 0 ||
3430 strcmp(nvpair_name(prop_err),
3431 zfs_prop_to_name(ZFS_PROP_REFQUOTA)) != 0) {
3433 * Skip the special case of, for example,
3434 * "refquota", errors on intermediate
3435 * snapshots leading up to a final one.
3436 * That's why we have all of the checks above.
3438 * See zfs_ioctl.c's extract_delay_props() for
3439 * a list of props which can fail on
3440 * intermediate snapshots, but shouldn't
3441 * affect the overall receive.
3443 (void) snprintf(tbuf, sizeof (tbuf),
3444 dgettext(TEXT_DOMAIN,
3445 "cannot receive %s property on %s"),
3446 nvpair_name(prop_err), zc.zc_name);
3447 zfs_setprop_error(hdl, prop, intval, tbuf);
3450 nvlist_free(prop_errors);
3453 zc.zc_nvlist_dst = 0;
3454 zc.zc_nvlist_dst_size = 0;
3455 zcmd_free_nvlists(&zc);
3457 if (err == 0 && snapprops_nvlist) {
3458 zfs_cmd_t zc2 = { 0 };
3460 (void) strcpy(zc2.zc_name, zc.zc_value);
3461 zc2.zc_cookie = B_TRUE; /* received */
3462 if (zcmd_write_src_nvlist(hdl, &zc2, snapprops_nvlist) == 0) {
3463 (void) zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc2);
3464 zcmd_free_nvlists(&zc2);
3468 if (err && (ioctl_errno == ENOENT || ioctl_errno == EEXIST)) {
3470 * It may be that this snapshot already exists,
3471 * in which case we want to consume & ignore it
3472 * rather than failing.
3474 avl_tree_t *local_avl;
3475 nvlist_t *local_nv, *fs;
3476 cp = strchr(zc.zc_value, '@');
3479 * XXX Do this faster by just iterating over snaps in
3480 * this fs. Also if zc_value does not exist, we will
3481 * get a strange "does not exist" error message.
3484 if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
3485 B_FALSE, &local_nv, &local_avl) == 0) {
3487 fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
3488 fsavl_destroy(local_avl);
3489 nvlist_free(local_nv);
3492 if (flags->verbose) {
3493 (void) printf("snap %s already exists; "
3494 "ignoring\n", zc.zc_value);
3496 err = ioctl_err = recv_skip(hdl, infd,
3503 if (ioctl_err != 0) {
3504 switch (ioctl_errno) {
3506 cp = strchr(zc.zc_value, '@');
3508 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3509 "most recent snapshot of %s does not\n"
3510 "match incremental source"), zc.zc_value);
3511 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3515 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3516 "destination %s has been modified\n"
3517 "since most recent snapshot"), zc.zc_name);
3518 (void) zfs_error(hdl, EZFS_BADRESTORE, errbuf);
3521 cp = strchr(zc.zc_value, '@');
3523 /* it's the containing fs that exists */
3526 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3527 "destination already exists"));
3528 (void) zfs_error_fmt(hdl, EZFS_EXISTS,
3529 dgettext(TEXT_DOMAIN, "cannot restore to %s"),
3534 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3537 recv_ecksum_set_aux(hdl, zc.zc_value, flags->resumable);
3538 (void) zfs_error(hdl, EZFS_BADSTREAM, errbuf);
3541 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3542 "pool must be upgraded to receive this stream."));
3543 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
3546 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3547 "destination %s space quota exceeded"), zc.zc_name);
3548 (void) zfs_error(hdl, EZFS_NOSPC, errbuf);
3551 (void) zfs_standard_error(hdl, ioctl_errno, errbuf);
3556 * Mount the target filesystem (if created). Also mount any
3557 * children of the target filesystem if we did a replication
3558 * receive (indicated by stream_avl being non-NULL).
3560 cp = strchr(zc.zc_value, '@');
3561 if (cp && (ioctl_err == 0 || !newfs)) {
3565 h = zfs_open(hdl, zc.zc_value,
3566 ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
3568 if (h->zfs_type == ZFS_TYPE_VOLUME) {
3570 } else if (newfs || stream_avl) {
3572 * Track the first/top of hierarchy fs,
3573 * for mounting and sharing later.
3575 if (top_zfs && *top_zfs == NULL)
3576 *top_zfs = zfs_strdup(hdl, zc.zc_value);
3584 if (!flags->nomount)
3585 err |= changelist_postfix(clp);
3586 changelist_free(clp);
3589 if (prop_errflags & ZPROP_ERR_NOCLEAR) {
3590 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3591 "failed to clear unreceived properties on %s"),
3593 (void) fprintf(stderr, "\n");
3595 if (prop_errflags & ZPROP_ERR_NORESTORE) {
3596 (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Warning: "
3597 "failed to restore original properties on %s"),
3599 (void) fprintf(stderr, "\n");
3602 if (err || ioctl_err)
3605 *action_handlep = zc.zc_action_handle;
3607 if (flags->verbose) {
3610 uint64_t bytes = zc.zc_cookie;
3611 time_t delta = time(NULL) - begin_time;
3614 zfs_nicenum(bytes, buf1, sizeof (buf1));
3615 zfs_nicenum(bytes/delta, buf2, sizeof (buf1));
3617 (void) printf("received %sB stream in %lu seconds (%sB/sec)\n",
3625 zfs_receive_impl(libzfs_handle_t *hdl, const char *tosnap,
3626 const char *originsnap, recvflags_t *flags, int infd, const char *sendfs,
3627 nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
3628 uint64_t *action_handlep, const char *finalsnap)
3631 dmu_replay_record_t drr, drr_noswap;
3632 struct drr_begin *drrb = &drr.drr_u.drr_begin;
3634 zio_cksum_t zcksum = { 0 };
3635 uint64_t featureflags;
3638 (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3641 if (flags->isprefix &&
3642 !zfs_dataset_exists(hdl, tosnap, ZFS_TYPE_DATASET)) {
3643 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified fs "
3644 "(%s) does not exist"), tosnap);
3645 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3648 !zfs_dataset_exists(hdl, originsnap, ZFS_TYPE_DATASET)) {
3649 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "specified origin fs "
3650 "(%s) does not exist"), originsnap);
3651 return (zfs_error(hdl, EZFS_NOENT, errbuf));
3654 /* read in the BEGIN record */
3655 if (0 != (err = recv_read(hdl, infd, &drr, sizeof (drr), B_FALSE,
3659 if (drr.drr_type == DRR_END || drr.drr_type == BSWAP_32(DRR_END)) {
3660 /* It's the double end record at the end of a package */
3664 /* the kernel needs the non-byteswapped begin record */
3667 flags->byteswap = B_FALSE;
3668 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC)) {
3670 * We computed the checksum in the wrong byteorder in
3671 * recv_read() above; do it again correctly.
3673 bzero(&zcksum, sizeof (zio_cksum_t));
3674 fletcher_4_incremental_byteswap(&drr, sizeof (drr), &zcksum);
3675 flags->byteswap = B_TRUE;
3677 drr.drr_type = BSWAP_32(drr.drr_type);
3678 drr.drr_payloadlen = BSWAP_32(drr.drr_payloadlen);
3679 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
3680 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
3681 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
3682 drrb->drr_type = BSWAP_32(drrb->drr_type);
3683 drrb->drr_flags = BSWAP_32(drrb->drr_flags);
3684 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
3685 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
3688 if (drrb->drr_magic != DMU_BACKUP_MAGIC || drr.drr_type != DRR_BEGIN) {
3689 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3690 "stream (bad magic number)"));
3691 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3694 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
3695 hdrtype = DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo);
3697 if (!DMU_STREAM_SUPPORTED(featureflags) ||
3698 (hdrtype != DMU_SUBSTREAM && hdrtype != DMU_COMPOUNDSTREAM)) {
3699 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3700 "stream has unsupported feature, feature flags = %lx"),
3702 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3705 if (strchr(drrb->drr_toname, '@') == NULL) {
3706 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "invalid "
3707 "stream (bad snapshot name)"));
3708 return (zfs_error(hdl, EZFS_BADSTREAM, errbuf));
3711 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) == DMU_SUBSTREAM) {
3712 char nonpackage_sendfs[ZFS_MAX_DATASET_NAME_LEN];
3713 if (sendfs == NULL) {
3715 * We were not called from zfs_receive_package(). Get
3716 * the fs specified by 'zfs send'.
3719 (void) strlcpy(nonpackage_sendfs,
3720 drr.drr_u.drr_begin.drr_toname,
3721 sizeof (nonpackage_sendfs));
3722 if ((cp = strchr(nonpackage_sendfs, '@')) != NULL)
3724 sendfs = nonpackage_sendfs;
3725 VERIFY(finalsnap == NULL);
3727 return (zfs_receive_one(hdl, infd, tosnap, originsnap, flags,
3728 &drr, &drr_noswap, sendfs, stream_nv, stream_avl, top_zfs,
3729 cleanup_fd, action_handlep, finalsnap));
3731 assert(DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
3732 DMU_COMPOUNDSTREAM);
3733 return (zfs_receive_package(hdl, infd, tosnap, flags, &drr,
3734 &zcksum, top_zfs, cleanup_fd, action_handlep));
3739 * Restores a backup of tosnap from the file descriptor specified by infd.
3740 * Return 0 on total success, -2 if some things couldn't be
3741 * destroyed/renamed/promoted, -1 if some things couldn't be received.
3742 * (-1 will override -2, if -1 and the resumable flag was specified the
3743 * transfer can be resumed if the sending side supports it).
3746 zfs_receive(libzfs_handle_t *hdl, const char *tosnap, nvlist_t *props,
3747 recvflags_t *flags, int infd, avl_tree_t *stream_avl)
3749 char *top_zfs = NULL;
3752 uint64_t action_handle = 0;
3753 char *originsnap = NULL;
3755 err = nvlist_lookup_string(props, "origin", &originsnap);
3756 if (err && err != ENOENT)
3760 cleanup_fd = open(ZFS_DEV, O_RDWR|O_EXCL);
3761 VERIFY(cleanup_fd >= 0);
3763 err = zfs_receive_impl(hdl, tosnap, originsnap, flags, infd, NULL, NULL,
3764 stream_avl, &top_zfs, cleanup_fd, &action_handle, NULL);
3766 VERIFY(0 == close(cleanup_fd));
3768 if (err == 0 && !flags->nomount && top_zfs) {
3770 prop_changelist_t *clp;
3772 zhp = zfs_open(hdl, top_zfs, ZFS_TYPE_FILESYSTEM);
3774 clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
3775 CL_GATHER_MOUNT_ALWAYS, 0);
3778 /* mount and share received datasets */
3779 err = changelist_postfix(clp);
3780 changelist_free(clp);
3783 if (zhp == NULL || clp == NULL || err)